VTT PUBLICATIONS 367

Master curve analysis of ductile to

brittle transition region fracture
toughness round robin data
The "EURO" fracture toughness curve

Kim Wallin
VTT Manufacturing Technology

TECHNICAL RESEARCH CENTRE OF FINLAND

ESPOO 1998
ISBN 951–38–5345–4 (soft back ed.)
ISSN 1235–0621 (soft back ed.)
ISBN 951–38–5346–2 (URL: http://www.inf.vtt.fi/pdf/)
ISSN 1455–0849 (URL: http://www.inf.vtt.fi/pdf/)
Copyright © Valtion teknillinen tutkimuskeskus (VTT) 1998

JULKAISIJA – UTGIVARE – PUBLISHER

Valtion teknillinen tutkimuskeskus (VTT), Vuorimiehentie 5, PL 2000, 02044 VTT
puh. vaihde (09) 4561, faksi (09) 456 4374
Statens tekniska forskningscentral (VTT), Bergsmansvägen 5, PB 2000, 02044 VTT
tel. växel (09) 4561, fax (09) 456 4374
Technical Research Centre of Finland (VTT), Vuorimiehentie 5, P.O.Box 2000, FIN–02044 VTT, Finland
phone internat. + 358 9 4561, fax + 358 9 456 4374

VTT Valmistustekniikka, Voimalaitosten materiaalitekniikka, Kemistintie 3, PL 1704, 02044 VTT

puh. vaihde (09) 4561, faksi (09) 456 7002
VTT Tillverkningsteknik, Material och strukturell integritet, Kemistvägen 3, PB 1704, 02044 VTT
tel. växel (09) 4561, fax (09) 456 7002
VTT Manufacturing Technology, Materials and Structural Integrity, Kemistintie 3, P.O.Box 1704, FIN–02044 VTT, Finland
phone international +358 9 4561, fax +358 9 456 7002

Technical editing Leena Ukskoski

Libella Painopalvelu Oy, Espoo 1998

Wallin, Kim. Master curve analysis of ductile to brittle transition region fracture toughness round
robin data. The "EURO" fracture toughness curve. Espoo 1998, Technical Research Centre of
Finland, VTT Publications 367. 58 p.
Keywords fractures (materials), steels, fracture strength, ductile brittle transition, brittleness,
statistical analysis, master curve method

Abstract
Brittle fracture in the ductile to brittle transition regime is connected with
specimen size effects and - more importantly- tremendous scatter of fracture
toughness, which the technical community is currently becoming increasingly
aware of. The size effects have the consequence that fracture toughness data
obtained from small laboratory specimens do not directly describe the fracture
behaviour of real flawed structures. Intensive research has been conducted in the
last decade in order to overcome these problems. Different approaches have
been developed and proposed, one of the most promising being the master curve
method, developed at VTT Manufacturing Technology.

For validation purposes, a large nuclear grade pressure vessel forging

22NiMoCr37 (A508 Cl.2) has been extensively characterised with fracture
toughness testing. The tests have been performed on standard geometry CT-
specimens having thickness 12.5 mm, 25 mm, 50 mm and 100 mm. The a/W-
ratio is close to 0.6 for all specimens. One set of specimens had 20% side-
grooves. The obtained data consists of a total of 757 results fulfilling the ESIS-
P2 test method validity requirements with respect to pre-fatigue crack shape and
the ASTM E-1921 pre-fatigue load. The master curve statistical analysis method
is meticuously applied on the data, in order to verify the validity of the method.
Based on the analysis it can be concluded that the validity of all the assumptions
in the master curve method is confirmed for this material.

3
 Preface
This work is part of the Measurements and Testing Programme of the European
Community Project: "Fracture Toughness of Steel in the Ductile to Brittle
Transition Regime" Contract No. MAT-CT-940080, co-ordinated by GKSS-
Forschungszentrum Geesthacht. The VTT participation in the project is part of
the national Material Degradation in Reactor Environment project (RAVA)
belonging to the Structural Integrity of NPP’s research programme (RATU2),
performed at VTT Manufacturing Technology and financed by the Ministry of
Trade and Industry in Finland (KTM), Technical Research Centre of Finland
(VTT), Radiation and Nuclear Safety Authority (STUK), and Finnish nuclear
power industry.

4
 Contents

Abstract 3

Preface 4

1. Introduction 7

2. Analysis method 8
2.1 Master curve distribution 8
2.2 Failure probability diagram 8
2.3 Rank probability 9
2.4 K0 estimation 10
2.5 Censoring 11
2.6 Thickness adjustment 12
2.7 Temperature dependence 12
2.8 T0 estimation 12
2.9 Kmin estimation 13
2.10 Lower shelf analysis 15

3. Preliminary analysis 16
3.1 Material 16
3.2 Test details 16
3.3 Yield stress 18
3.4 Raw data 18
3.5 Empirical analysis 21

4. Master curve analysis 26

4.1 T = -154°C 26
4.2 T = -110°C 29
4.3 T = -91°C 29
4.4 T = -60°C 31
4.5 T = -40°C 35
4.6 T = -20°C 36
4.7 T = -10°C 40
4.8 T = 0°C 41
4.9 T = +20°C 43

5
5. Synthesis analysis 46
5.1 Lower shelf behaviour 46
5.2 Size effect of K0 48
5.3 Temperature dependence of K0 49
5.4 Validity of fixed Kmin 51
5.5 Multi-temperature master curve analysis 52

6. Summary and conclusions 55

References 57

6
 1. Introduction
The European Structural Integrity Society (ESIS) and the International
Standardisation Organisation (ISO) are currently drafting test standards for
characterising the fracture toughness of metals which are supposed to be basis
for CEN standards. The drafts include also the characterisation of the transition
from ductile to brittle fracture of steels.

However, this transition region is connected with specimen size effects and −
more importantly − tremendous scatter of fracture toughness, which the
technical community is currently becoming increasingly aware of.

The size effects have the consequence that fracture toughness data obtained
from small laboratory specimens do not directly describe the fracture behaviour
of real flawed structures. Intensive research has been conducted in the last
decade in order to overcome these problems. Different approaches have been
developed and proposed, one of the most promising being the master curve
method, developed at VTT Manufacturing Technology. The master curve
method, provides a description for the fracture toughness scatter, size effect and
temperature dependence both for the transition region as well as the lower shelf.
The master curve method has already led to an American testing and analysis
standard ASTM E1921-98 and it is an obvious candidate for European
standardisation. In order to verify the validity of the master curve method and
other proposed methods, a sufficiently large data set which shows continuously
the scatter and the size effects from the lower shelf to the upper shelf of a single
material is needed. Such a data has not existed until now.

In 1995 a project entitled "Fracture Toughness of Steel in the Ductile to Brittle

Transition Regime" was launched under the "Measurement and Testing
Programme of the European Community". In the project, more than 750,
specimens of four different sizes have been tested at different temperatures to
produce a sufficiently large data set for the validation of the different statistical
methods. In this report, the data from the project is meticulously analysed with
the, VTT based, master curve method and the validity of the different aspects of
the method is verified, both separately and in combination.

7
 2. Analysis method

2.1 Master curve distribution

The master curve cumulative failure probability distribution has the form (Eq. 1)
[1].

 K −K  
4

Pf = 1 − exp −  JC min
  (1)
  K 0 − K min  

2.2 Failure probability diagram

The single temperature data is presented in the “original” master curve failure
probability diagram which produces a linear presentation of the master curve
cumulative failure probability distribution (Eq. 2, Fig. 1) [1].

1/ 4
 1  K Jc − K min
ln  = (2)
 1 − Pf  K 0 − K min

Nearly symmetric
95 %
1.5 rank probability
confidence bounds
 1 1/4

5%
=> K0

 1−P 
f

1.0
Clear description
ln

of Kmin


0.5

0.0 Linear toughness

0 50 100 150 200 250 300 350
representation
KIC [MPa√m]

Figure 1. Master curve failure probability diagram.

8
Compared to a normal Weibull probability diagram, the master curve failure
probability diagram offers a better visualisation of the data with respect to the
master curve distribution, as indicated in Fig. 1.

2.3 Rank probability

When the data is plotted into the failure probability diagram, it must be ordered
by rank and designated rank probabilities. The weakness with the rank
probability estimates are that they are not measured values, but estimates of
cumulative probability based on order statistics. Each data point corresponds to
a certain cumulative failure probability with a certain confidence. This can be
expressed in a mathematical form, using the binomial distribution, as [2]

i n! j −1 n − j +1
z' = 1 − ∑ ⋅P ⋅ (1 − P ) (3)
j = 1 ( j − 1)!⋅ (n − j + 1)!
rank rank

where z’ is the probability that the rank estimate corresponds to the cumulative
probability Prank, n is the number of points and i is the rank number. Eq. 3 can
be used to calculate rank confidence estimates like the ones presented in Fig. 1.
Usually people prefer to use simple approximations of the median rank
probability estimate (z’ = 0.5). Three common estimates of the median rank
probability are [2]

i − 0.5
Prank = (4a)
n

i
Prank = (4b)
n +1

i − 0.3
Prank = (4c)
n + 0.4

The three approximations are compared with the outcome of Eq. 3 in Fig. 2. It
can be seen that Eq. 4c is clearly the best estimate of the median rank
probability. Thus, Eq. 4c is used for the presentation of the single temperature
round robin data.

9
 1.0

i
0.8 1
= n+
i - Prank ⋅ n

i − 0.5
k
0.6 Prank =
P ran
n
0.4
i − 0.3
Prank =
n + 0 .4
0.2

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Prank

Figure 2. Comparison of different estimates of median rank probability (lines)

with binomial theory estimate (circles).

2.4 K0 estimation

The estimation of the normalisation toughness K0 (corresponding to 63.2%

failure probability) is based on the randomly censored Maximum Likelihood
expression (Eq. 5) [3].

 n 4 1 / 4

(
 ∑ K i − K min 

)
K0 =  i = 1  + K min (5)
 n 
 ∑ δ − 1 + ln 2 
 i = 1i 
 

where the censoring parameter δi is 1 for uncensored and 0 for censored data
and the limiting fracture toughness Kmin is fixed as 20 MPa√m.
n
The accuracy of the K0 estimate will be a function of uncensored data r = ∑ δ i .
i =1

10
The standard deviation of the bias corrected MML estimate is approximately [3]

0.28 ⋅ ( K 0 − K min )
σ K − K ≈ (6)
0 min r

An important feature, of the MML estimate, to note is that, the accuracy of the
K0 is not affected by the degree of censoring, i.e. it is not affected by the n/r
ratio. Nor is it affected by the nature of censoring (random, upper end, lower
end, etc.) It is only a function of the number of uncensored data (r). This can
easily be verified by e.g. Monte Carlo simulations, using different censoring
criteria and degrees of censoring. This property makes the MML procedure
extremely valuable. In principle, not a single test result (fulfilling the
requirements related to fatigue precracking, crack shape and test performance)
needs to be omitted from the analysis. Any invalid data can be given the
toughness value corresponding to the validity criteria and be treated as non-
failures.

2.5 Censoring

Two levels of censoring are applied. First, all data referring to “non-cleavage”
(ductile end of test) are prescribed δi = 0. Second, all data violating the
specimen size validity criterion (Eq. 7) were designated the toughness value
corresponding to the validity criteria and given δi = 0.

The specimen size validity criterion (taken same as in the ASTM master curve
standard [4]) has the form

b ⋅σ y ⋅ E
K JC ≤
( )
(7)
M ⋅ 1−υ2

where σy is yield strength, E is the modulus of elasticity, the controlling

dimension is the ligament size b and the size criterion constant M = 30.

11
 2.6 Thickness adjustment

For the comparison of different size specimen data and for the calculation of the
master curve transition temperature T 0 all data is thickness adjusted to the
reference flaw length (thickness) B0 = 25 mm with Eq. 8 [5].

1/ 4
 B
K 25mm = K min + ( K JC − K min ) ⋅   (8)
 B0 

The thickness B refers to the nominal thickness, regardless of side-grooving. I.e.

the 25 mm thick side-grooved specimens tested at -20°C, having a net thickness
of 20 mm, are treated as being 25 mm thick.

2.7 Temperature dependence

In the master curve concept, a constant temperature dependence for cleavage

fracture toughness is assumed (Eq. 9) [6, 7].

K 0 = 31 + 77 ⋅ exp{0.019 ⋅ (T − T0 )} (9)

The transition temperature T0 corresponds to the temperature where the mean

(median) fracture toughness for a 25 mm thick specimen has the value 100
MPa√m.

2.8 T0 estimation

Two methods of estimating the master curve transition temperature T0 are used.
First, for the single temperature data, T0 is calculated from the size adjusted K0
values using Eq. 9. Second, for the multi-temperature data, T0 is estimated from
the size adjusted KJc data using a multi-temperature randomly censored
maximum likelihood expression (Eq. 10) [8].

12
 n {
δ ⋅ exp 0.019 ⋅ T − T
i i 0 [ ]}

K
n  ICi
4
{ [
− 20 ⋅ exp 0.019 ⋅ T − T
 i 0 ]} (10)
{ ]}
∑ − ∑ =0
i = 1 11 + 77 ⋅ exp 0.019 ⋅ T − T
i 0 [ 
+ ⋅ { [
i = 1 11 77 exp 0.019 T T 
−
i 0 ]}
5

The transition temperature T0 is solved iteratively from Eq. 10.

Since the accuracy of the K0 estimate is proportionally constant with respect to

K0 (Eq. 6), the accuracy of the T0 will be a function of K0. Low K0 leads to large
uncertainty in T0 and high K0 leads to small uncertainty in T0 (Fig. 3).

40
5%
20
⋅√ r [ C]

0
0

T0x% = T0 + ∆T0x%
-20
CONF.

-40 95 %
∆ T0

-60 r = number of cleavage

results
-80

-150 -100 -50 0 50 100

0
T-T0 [ C]

Figure 3. 5% and 95% confidence bounds for MML T0 estimate.

2.9 Kmin estimation

In addition to the standard maximum likelihood expression (using a fixed value

of Kmin [20 MPa√m]), also the more complicated procedure of fitting both K0
and Kmin was applied. The randomly censored maximum likelihood expression
for fitting both K0 and Kmin is (Eq. 11) [2]

13
 δi
( )
n n

∑ ⋅ ∑ K i − K m in
4

3 i =1 K i − K m in i =1
⋅ − 1 = 0 ⇒ estim ate of K m in
∑ δ ⋅ ∑ (K )
n n
4 3
i i − K m in
i =1 i =1
1/ 4
 n 4

 ∑ K i − K m in
 i =1
( ) 
 (11)
K0 =  n  + K m in
 ∑ δ i − 1 + ln 2 
 i =1 
n

∑δ i ⋅ K m in − K 0
i =1
K m in = n

∑δ i −1
i =1

where the censoring parameter δi is 1 for uncensored and 0 for censored data.
Eq. 11 includes a small bias correction for both K0 and Kmin.

The accuracy of the Kmin estimate is a function of number of uncensored data

and the ratio between K0/Kmin (Fig. 4) [9].

0.5 95 %
min
min

0
−K
K −K
min

-0.5
0

5%
K


-1

-1.5

1 6 10 100
NUMBER OF SIGNIFICANT TESTS

Figure 4. Accuracy of K min MML estimate.

14
 2.10 Lower shelf analysis

The above derivations are based on the assumption that the probability of
cleavage initiation is less than unity. This is normal for configurations like plain
and notched specimens and cracked specimens in cases where initiation is
sufficiently difficult. For material conditions where initiation is simple, the
probability of cleavage initiation in the case of a crack may become unity. This
can occur on the so called “lower shelf” of the material. Essentially it means
that all possible initiation sites are activated and initiation occurs as soon as the
crack is loaded, making the initiation event independent of the load level (and
subsequently independent of specimen thickness). Thus, in the case of a crack,
the lower shelf toughness may be controlled purely by the probability of
propagation. For plain and notched configurations, however, the probability of
initiation will still be a function of load level even on the lower shelf. Therefore,
a simple correlation between notched and cracked configurations may not be
possible for the lower shelf material conditions.

The three data sets corresponding to test temperature -1540C, are additionally
analysed with a special equation developed for the propagation controlled lower
shelf behaviour of fracture toughness.

The equation has the form ( Eq. 12) [10]

 K  K  K  K K  2 3 
 min4  I  I  min4 1  min4   (12)
P
f4LS
= 1 − exp- ⋅ − 3 ⋅ ln  − 3⋅ K +   + 2 
 Ke4 K
 min4
K
 min4  1
2  K
I  


where Kmin4 is the lower shelf limiting minimum fracture toughness and Ke4 is a
normalisation fracture toughness.

The biggest difference between the lower shelf expression and the transition
region expression is that the lower shelf expression does not predict a statistical
size effect.

15
 3. Preliminary analysis

3.1 Material

All specimens were extracted from a single segment of a large nuclear grade
pressure vessel forging 22NiMoCr37 (A508 Cl.2) so that the crack front was
located in the region ¼T–½T which had been found to be “homogeneous” in the
preliminary investigations performed by GKSS [11]. The sectioning diagram is
presented in Fig. 5.

Due to available time and resources, the specimens were not randomly
distributed throughout the forging, but instead the large segment was divided
into smaller sections from which different size specimens were manufactured
and tested at one temperature. All the larger specimens could not be extracted
form the same section as the small specimens, but an effort was made to take
them comparatively close from the small specimen sections.

3.2 Test details

The fracture toughness tests were performed on standard geometry CT-

specimens having thickness 12.5 mm, 25 mm, 50 mm and 100 mm. The a/W-
ratio was close to 0.6 for all specimens.

The fracture toughness tests were performed essentially in accordance with the
ESIS P2 procedure without performing a crack growth correction on the J-
integral. The main deviation from the ESIS P2 procedure occurred for the pre-
fatigue of the specimens, which was performed in line with the ASTM E-1921
standard. One set of 25 mm thick specimens tested at -20oC was side-grooved
20% (10% + 10%). All specimens were loaded either until occurrence of brittle
fracture or until the load drop due to ductile tearing equalled 20% of the
maximum load.

16
 SX5 SX6 ½T, 1T, 2T, -154°C
2T, -154°C
SX1 1T, -10°C
4T, +20°C
SX7 SX8 ½T, 1T, 2T, -90°C
2T, -91°C 1T, 0°C, 4T, -91°C
4T, +20°C

2T, -60°C SX9 SX10 ½T, 1T, 2T, -60°C

(½T, -60°C) 1T, 0°C
4T, +20°C
SX11 SX12 ½T, 1T, 2T, -40°C
2T, -40°C 4T, -20°C
SX2 4T, -91°C
2T, -154°C SX13 SX14 1T0SG, -20°C
2T, -91°C 2T, +20°C 4T, -20°C, 1T, +20°C
2T, -60°C 2T, +20°C
(½T, -60°C)
2T, -40°C SX15 SX16 ½T, 1T, 2T, -20°C
2T, -20°C 2T, -20°C 4T, -20°C
2T, 0°C 4T, -91°C
2T, +20°C SX17 SX18 ½T, 1T, 2T, 0°C
4T, +20°C 2T, 0°C 4T, -20°C

SX20 SX21 SX22 SX23

SX19 SX24
SX3 4T, -91°C
4T, 0°C 4T, -20°C 4T, -20°C 4T, -91°C 4T, -20°C
4T, +20°C 4T, 0°C
4T, 0°C 4T, 0°C

T
L
SX25 SX4
4T, -91°C
4T, -20°C

kw981.dsf

Figure 5. Sectioning diagram of 22NiMoCr37.

The measured J-integral values were transformed into KJc values with the
equation

J ⋅ 206 GPa
K Jc = (13)
0.91

i.e. a constant modulus of elasticity and plane strain conditions were assumed
for all temperatures.

17
A more detailed description of the project background and test performance is
presented elsewhere [11].

3.3 Yield stress

The measured yield stress values were fitted by an exponential temperature

dependence expression, the result of which is presented in Fig. 6.

22NiMoCr37
750

700 σYT = 450 + 1294 ⋅ exp(-0.0147⋅T[K])

650
σYT [MPa]

600

550

500

450
-150 -100 -50 0 50
0
T [ C]

Figure 6. Yield stress temperature dependence.

3.4 Raw data

All the fracture toughness results, fulfilling the initial crack front straightness
and pre-fatigue criteria, are presented in Fig. 7. The data is unadjusted “raw”
data. In addition to the fracture toughness, also the amount of ductile tearing
preceding brittle fracture was measured optically. The resulting multi-specimen
KJc-∆a-curve is presented in Fig. 8. In Fig. 8, all specimens failing by cleavage
have been included. Thus also data referring to values beyond maximum load
are included. Only beyond the values shown in Fig. 8 the KJc-∆a-curves start to
show a clear size dependence (small specimens yield lower curves).

18
 22NiMoCr37 N = 757
1800

1600 C D
B = 12.5 mm
1400
B = 25 mm
KJc [MPa√m]

1200 B = 50 mm
B = 100 mm
1000

800

600

400

200

0
-160 -140 -120 -100 -80 -60 -40 -20 0 20 40
0
T [ C]

Figure 7. “Raw” fracture toughness data. Filled points refer to ductile end of
test values.

The KJ values corresponding to ductile load maximum are presented in Fig. 9. A

remarkable resemblance between Figs 8 and 9 is clear. The two plots seem to
represent the same tearing resistance curve. This indicates that the amount of
crack growth at maximum load is directly proportional to the specimen ligament
size (not thickness of specimen) and the maximum load value corresponds only
to a specific location on the tearing resistance curve. The amount of ductile
tearing at maximum load will also be a function of the materials tearing
resistance and strain hardening, but for a single material and specimen type, the
controlling parameter will be the ligament size. This result gives a clear
interpretation of the relevance of the ductile maximum load fracture toughness.

19
 22NiMoCr37
1200

1000 KJ = 439⋅∆a
0.36
KJo [MPa√m]

800

600

B = 12.5 mm
400 B = 25 mm
B = 50 mm
200 B = 100 mm

0
0 2 4 6 8 10 12
∆a [mm]

Figure 8. Multi-specimen (cleavage fracture) KJ-∆a-curve showing no effect of

specimen size.

22NiMoCr37
1000

0.33
KJCmax = 500⋅{(W-a)/25}
800 load
[MPa√m]

600
load
KJmax

400

200

0
0 10 20 30 40 50 60 70 80 90
W-a [mm]

Figure 9. KJ at ductile maximum load as a function of specimen ligament size.

20
Some testing standards which recognise the maximum load fracture toughness
(e.g. BS 7448) demand the use of full thickness specimens, but allow specimens
with different ligament size. This is clearly wrong. For standard type bend
specimens, the controlling dimension of the specimen is the ligament and not
the thickness. The smaller the specimen ligament is the smaller the maximum
load fracture toughness will be. Therefore, the demand for full thickness
specimens is inappropriate.

3.5 Empirical analysis

Before performing a detailed master curve analysis of the results, a simple

empirical analysis was performed to check the overall behaviour of the data.
First the data for the individual specimen sizes was plotted in logarithmic co-
ordinates to check trends in temperature dependence and scatter (Fig. 10). A
Least Square Fit was performed on the data to determine an approximate
estimate of the mean fracture toughness behaviour (solid lines in Fig. 10). At
this point no distinction was made between failure and non-failure data. Also,
deterministic “lower bound” estimates were developed by drawing lower
envelope lines to the data (dashed lines in Fig. 10). The behaviour of the four
specimen sizes is very similar. In all cases the proportional scatter is seen to
increase when going to higher temperatures. This can have three explanations:
1
at higher temperatures there may be a loss of constraint, increasing the scatter,
2
there may be a lower limiting fracture toughness Kmin causing the proportional
scatter in terms of KJc to be a function of the KJc/Kmin ratio (Eq. 6) or 3the master
curve assumption of the scatter behaviour may be wrong. However, the
similarity between the three large data sets (12.5 mm, 25 mm and 50 mm)
indicate a specific fixed scatter dependence (difference between mean and lower
bound estimates is similar). The difference between mean and lower bound
estimates for the 100 mm thick specimens is less, but this is directly attributable
to the smaller number of tested specimens (the deterministic lower bound
corresponds to a higher probability level). Explanations one and two appear
most likely, operating separately or in combination.

The mean and lower bound curves are compared in Fig. 11. The mean curves
show a clear size effect. The effect of increasing the specimen size is essentially
to shift the curves higher in temperature. The shape of the mean curves appears

21
practically unaffected by specimen size. A doubling of specimen size seems
equivalent to a constant temperature shift of approximately 9°C. Specimen size
effects are less clear for the lower bound curves, but this may to some extent be
due to their higher inherent uncertainty. Also the fact that the 100 mm thickness
lower bound seem to coincide with the 12.5 mm thickness lower bound is (at
least partly) due to the lesser number of 100 thickness specimens. Otherwise the
results show the expected trend of decreasing lower bound toughness with
increasing specimen size.

B = 12.5 mm N = 268 1000

B = 25 mm N = 224
1000 CLEAVAGE CLEAVAGE
DUCTILE DUCTILE
KJc [MPa√m]

KJc [MPa√m]

MEAN CURVE MEAN CURVE

100 100

LOWER BOUND LOWER BOUND

ENVELOPE ENVELOPE

10 10
-160 -140 -120 -100 -80 -60 -40 -20 0 20 40 -160 -140 -120 -100 -80 -60 -40 -20 0 20 40
0 0
T [ C] T [ C]
22NiMoCr37 N = 757
B = 50 mm N = 210 B = 100 mm N = 55
1000 CLEAVAGE 1000
CLEAVAGE
DUCTILE DUCTILE
KJc [MPa√m]

KJc [MPa√m]

MEAN CURVE MEAN CURVE

100
100
LOWER BOUND
LOWER BOUND ENVELOPE
ENVELOPE
10
-160 -140 -120 -100 -80 -60 -40 -20 0 20 40 -160 -140 -120 -100 -80 -60 -40 -20 0 20 40
0 0
T [ C] T [ C]

Figure 10. Individual specimen size fracture toughness results expressed in

logarithmic co-ordinates. Solid lines are LSF mean estimates using all
specimens (including non-failures) and dashed lines are “eye ball”
deterministic lower bound estimates.

22
 22NiMoCr37 N = 757
600
MEAN CURVES
500
B = 12.5 mm N = 268
B = 25 mm N = 224
400
KJc [MPa√m]

B = 50 mm N = 210
B = 100 mm N = 55
300

200

100
LOWER BOUND
ENVELOPES
0
-150 -125 -100 -75 -50 -25 0 25

0
T [ C]

Figure 11. Comparison of LSF mean and “eye ball” lower bound fracture
toughness estimates.

The simple mean estimates were compared with a master curve prediction of the
median fracture toughness using a T0 of -95°C (Fig. 12a).

The LSF mean curve estimates and the master curve predictions are clearly
similar. The size effect is practically identical and also the curve shapes are
similar. Small differences in the shape are to be expected, since the LSF mean
predictions used a simple exponential function, extrapolating down to zero
fracture toughness. The master curve assumes a slightly more advanced form of
exponential function utilising an absolute lower shelf value of fracture
toughness. Therefore, at low temperatures, the master curve prediction will be
higher than the LSF mean estimate and, consequently, the reverse will be the
case at high temperatures since the master curves were located to describe the
average behaviour of the LSF estimates.

The comparison was repeated for the deterministic lower bound curves. They
were compared with the 1% (and 5% for the 100 mm specimens) predictions of
the master curve (Fig. 12b). In this case, the master curve seems to predict a
larger size effect than shown by the deterministic curves. Actually the 12.5 mm,

23
25 mm and 100 mm lower bound curves are almost identical. This finding may
be somewhat misleading. The number of 100 mm specimens is less than ¼ of
the small specimens. Thus, the 100 mm deterministic lower bound is likely to
correspond to a higher probability level (4 times) than for the others. The fact
that the 5% master curve describes the 100 mm lower bound quite well indicates
that 1% (or slightly higher) probability master curves should describe the
smaller specimen lower bounds. Considering the uncertainty of deterministic
lower bound curves, the results appear quite promising. No clear discrepancies
with the master curve assumption is seen.

The empirical preliminary analysis indicates promise of having success with a

detailed analysis of the data applying the master curve analysis method
described in chapter 1. Thus, such an analysis is attempted next.

22NiMoCr37 N = 757
800

700 MEAN FIT

MASTER CURVE
600
KJc [MPa√m]

0
T0 = -95 C
500
B = 12.5 mm
400 B = 25 mm
B = 50 mm
300 B = 100 mm

200

100

0
-160 -140 -120 -100 -80 -60 -40 -20 0 20
0
T [ C]

Figure 12a. Comparison of LSF mean estimates and the master curve median
predictions using T0 = -95°C.

24
 22NiMoCr37 N = 757

LOWER BOUND
200 1 % MASTER CURVE
5 % MASTER CURVE (100 mm)
KJc [MPa√m]

0
T0 = -95 C
150
B = 12.5 mm
B = 25 mm
100
B = 50 mm
B = 100 mm

50

0
-160 -140 -120 -100 -80 -60 -40 -20 0 20
0
T [ C]

Figure 12b. Comparison of deterministic lower bound estimates and the master
curve 1% (& 5%) lower bound predictions using T0 = -95°C.

25
 4. Master curve analysis
The master curve analysis is first discussed separately for each test temperature
and subsequently an overall synthesis is made. The statistical size adjustment
(Eq. 8) is not performed for the analysis of the different test temperatures, but
only for the synthesis analysis.

4.1 T = -154°C

The temperature -154°C corresponds to the lower shelf region of fracture

toughness. From the fracture surfaces this could be seen as a lack of single
initiation sites. At this temperature the lower shelf behaviour depicted in Fig. 13
was seen for all but one specimen of 12.5 mm thickness. At higher temperatures
the transition region (Fig. 13) behaviour became more and more dominant,
starting with the smaller specimen sizes. This behaviour is well in accordance
with the master curve assumption of the lower shelf and transition region
fracture micromechanisms. The initiation site behaviour is further discussed in
the synthesis section.

TRANSITION LOWER
REGION SHELF
Crack propagation

Figure 13. Typical cleavage fracture surfaces for fatigue precracked specimens.

The non-size-adjusted -154°C test results are presented in Figs 14–16 for
specimen thickness 12.5 mm, 25 mm and 50 mm, respectively. All three cases
are strikingly similar. The standard master curve description, with a fixed Kmin,

26
does not provide a very good description of the data, but this is expected, since
the standard master curve is not supposed to work on the lower shelf. A slightly
better description of the data is obtained with the master curve description,
fitting also Kmin, but clearly the best result is obtained with the expression
developed specifically for the lower shelf (Eq. 12). The lower shelf expression
correctly models the curvature seen in the data when plotted in master curve
failure probability diagram co-ordinates. Hardly any size effect is visible in the
results and, overall, there are no significant differences between the three data
sets. The success of the lower shelf expression in describing the data and the
fact that the fracture surfaces lack single initiation sites are strong evidence for
the validity of the master curve assumption regarding the fracture mechanism on
the lower shelf.

1.4 1.4
CLEAVAGE CLEAVAGE
1.2 DUCTILE 95 % 1.2 95 %
DUCTILE
K0 = 43 MPa√m
1 4
1/ 4

K0 = 43 MPa√m
1.0 1.0
/

Kmin = 20 MPa√m Kmin = 13 MPa√m

f

f

 ln 1 
 1− P 

 ln 1 
 1−P 

0.8 0.8 5%
5%
0.6 0.6


0.4 0.4

0.2 0.2
FIXED Kmin FITTED Kmin
0.0 0.0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
KJC [MPa√m] KJC [MPa√m]

1.4

1.2 CLEAVAGE 95 %
DUCTILE
1/ 4

Kmin = 20.2 MPa√m

1.0
Ke = 1.11 MPa√m
5% 22NiMoCr37




f

0.8
T = -1540C
 1−P
 ln 1

0.6
B = 12.5 mm


0.4

0.2
Lower Shelf
0.0
0 10 20 30 40 50 60
KJC [MPa√m]

Figure 14. Failure probability diagram for 12.5 mm thick specimens at T =

-154°C.

27
 1.4 1.4

1.2 CLEAVAGE 95 % 1.2 CLEAVAGE 95 %

DUCTILE DUCTILE

 1/ 4
1/ 4

1.0 K0 = 42 MPa√m 1.0 K0 = 42 MPa√m

f

f

 ln 1 
 1− P 

 ln 1 
 1−P 
Kmin = 20 MPa√m Kmin = 14 MPa√m
0.8 0.8
5% 5%
0.6 0.6



0.4 0.4

0.2 0.2
FIXED Kmin FITTED Kmin
0.0 0.0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
KJC [MPa√m] KJC [MPa√m]

1.4
CLEAVAGE 95 %
1.2 DUCTILE
Kmin = 20.8 MPa√m
1/ 4

1.0
Ke = 0.84 MPa√m 5%
22NiMoCr37
f



f

0.8
 1−P
 ln 1

0.6 T = -1540C
B = 25 mm


0.4

0.2
Lower Shelf
0.0
0 10 20 30 40 50 60
KJC [MPa√m]

Figure 15. Failure probability diagram for 25 mm thick specimens at T =

-154°C.

1.4 1.4
95 % CLEAVAGE 95 %
1.2 CLEAVAGE 1.2 DUCTILE
DUCTILE
 1/ 4
1/ 4

K0 = 39 MPa√m
1.0 K0 = 40 MPa√m 1.0
Kmin = 14 MPa√m
f

f

Kmin = 20 MPa√m

 ln 1 
 1− P 

 ln 1 
 1−P 

5%
0.8 0.8
5%
0.6 0.6


0.4 0.4

0.2 0.2
FIXED Kmin FITTED Kmin
0.0 0.0
0 10 20 30 40 50 60 0 10 20 30 40 50 60
KJC [MPa√m] KJC [MPa√m]

1.4
95 %
CLEAVAGE
1.2
DUCTILE
Kmin = 21.2 MPa√m
1/ 4

1.0 5%
Ke = 0.53 MPa√m 22NiMoCr37





0.8
f
 1−P
 ln 1

0.6
T = -1540C
B = 50 mm


0.4

0.2
Lower Shelf
0.0
0 10 20 30 40 50 60
KJC [MPa√m]

Figure 16. Failure probability diagram for 50 mm thick specimens at T =

-154°C.

28
 4.2 T = -110°C

The 55 tests at -110°C were performed as part of the materials pre-

characterisation and therefore only one specimen size was used (12.5 mm
thickness). Only 27% of the specimens lacked a single initiation site and
therefore the basic master curve expression should work. The results are
presented in Fig. 17. Using a fixed Kmin or fitting does not make much
difference. The fitted Kmin is very close to the fixed value. Considering the large
number of specimens in this data set, the result constitutes a strong validation of
the assumed master curve scatter and Kmin.

22NiMoCr37 T = -1100C B = 12.5 mm

1.4 1.4
CLEAVAGE
1.2 DUCTILE CLEAVAGE
95 % 1.2
DUCTILE 95 %
K0 = 88 MPa√m
 1/ 4

K0 = 88 MPa√m
 1/ 4

1.0 Kmin = 20 MPa√m 1.0





Kmin = 18 MPa√m



f
 1−P

0.8 0.8
f
 ln 1

 1−P

5% 5%
 ln 1

0.6 0.6


0.4 0.4

0.2 FIXED Kmin 0.2 FITTED Kmin

0.0 0.0
0 20 40 60 80 100 120 0 20 40 60 80 100 120
KJC [MPa √m] KJC [MPa √m]

Figure 17. Failure probability diagram for 12.5 mm thick specimens at T =

-110°C.

4.3 T = -91°C

All four specimen sizes were tested at -91°C and the non-size-adjusted results
are presented in Figs 18–21. With the exception of the 100 mm specimens more
than 50% of the specimens showed single initiation sites, so the weakest link
based master curve expression should work. For the 100 thick specimens only
one specimen showed a clearly recognisable single initiation site, so in this case
lower shelf behaviour could be expected. This was not however supported by
the fracture toughness behaviour (Fig. 21) which follows the master curve
prediction very well. The discrepancy may be due to the difficulty of
recognising initiation sites correctly on large fracture surfaces. Overall,
however, the results show the expected size effect and scatter. The 50 mm thick

29
specimens do not quite confirm the expected scatter, but this may be due to a
macroscopic inhomogeneity in the material used for the 50 mm specimens
(verified for tests at -60°C) or to a statistical sampling effect (30 specimens is
still small sample statistics). Most importantly, the standard master curve is
shown to produce a realistic or conservative description of the data. The fitted
Kmin values are in all cases close to the fixed standard master curve value of Kmin
= 20 MPa√m.

22NiMoCr37 T = -910C B = 12.5 mm

1.4 1.4
CLEAVAGE CLEAVAGE
1.2 DUCTILE 1.2 DUCTILE
K0 = 122 MPa√m 95 % K0 = 122 MPa√m 95 %
 1/ 4

1.0 1.0
 1/ 4

Kmin = 20 MPa√m Kmin = 20 MPa√m





5% 5%




0.8 0.8
f
 1−P

f
 ln 1

 1−P
 ln 1

0.6 0.6


0.4 0.4

0.2 FIXED Kmin 0.2

FITTED Kmin
0.0 0.0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
KJC [MPa √m] KJC [MPa √m]

Figure 18. Failure probability diagram for 12.5 mm thick specimens at T =

-91°C.

22NiMoCr37 T = -910C B = 25 mm
1.4 1.4
CLEAVAGE CLEAVAGE
1.2 DUCTILE 95 % 1.2 DUCTILE 95 %
K0 = 112 MPa√m K0 = 111 MPa√m
1 4

1.0 Kmin = 20 MPa√m 1.0

1 4
/

Kmin = 15 MPa√m
/




f
 ln 1 
 1−P 

0.8 0.8
f
 1−P

5%
 ln 1

5%
0.6 0.6


0.4 0.4

0.2 FIXED Kmin 0.2

FITTED Kmin
0.0 0.0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
KJC [MPa √m] KJC [MPa √m]

Figure 19. Failure probability diagram for 25 mm thick specimens at T =

-91°C.

30
 22NiMoCr37 T = -910C B = 50 mm
1.4 1.4

1.2 FIXED Kmin 1.2 FITTED Kmin

1/ 4

1.0 95 % 1.0 95 %

1/ 4
5%









0.8 0.8
f
 1− P

5%

f
 ln 1

 1− P
 ln 1
0.6 CLEAVAGE 0.6
CLEAVAGE


DUCTILE


DUCTILE
0.4 0.4
K0 = 101 MPa√m K0 = 101 MPa√m
0.2 Kmin = 20 MPa√m 0.2 Kmin = 15 MPa√m

0.0 0.0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
KJC [MPa √m] KJC [MPa √m]

Figure 20. Failure probability diagram for 50 mm thick specimens at T =

-91°C.

22NiMoCr37 T = -910C B = 100 mm

1.4 1.4

1.2 95 % 1.2 95 %
FIXED K min
FITTED Kmin
1/ 4

1.0 1.0
1/ 4










0.8 0.8
f
 1− P

f
 ln 1

 1− P
 ln 1

5% 5%
0.6 0.6
CLEAVAGE CLEAVAGE


DUCTILE DUCTILE
0.4 0.4
K0 = 87 MPa√m K0 = 88 MPa√m
0.2 Kmin = 20 MPa√m 0.2 Kmin = 27 MPa√m

0.0 0.0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
KJC [MPa √m] KJC [MPa √m]

Figure 21. Failure probability diagram for 100 mm thick specimens at T =

-91°C.

4.4 T = -60°C

Three smallest specimen sizes were tested at -60°C and the non-size-adjusted
results are presented in Figs 22–24. Only for the 50 mm thick specimens, there
were 40% specimens showing lack of single initiation sites. Thus the basic
master curve expression should work. All sets follow the predicted distribution
quite well. At this temperature the 12.5 mm thick specimens start to reach
fracture toughness values in excess of the size requirement (Eq. 7). A loss of
specimen constraint should be visible as a deviation towards the right from the
straight line. Such a deviation does not occur before the M = 30 line, thus
providing validation for the standard size requirement.

31
 22NiMoCr37 T = -600C B = 12.5 mm
1.4 1.4
CLEAVAGE CLEAVAGE
1.2 DUCTILE 95 %
1.2 DUCTILE
95 %
K0 = 169 MPa√m K0 = 168 MPa√m
1/ 4

1.0 1.0

1/ 4
Kmin = 20 MPa√m Kmin = 2 MPa√m







 ln 1 
 1−P 
0.8 0.8
f
 1− P

5%

f
 ln 1

5%
0.6 0.6



0.4 0.4

0.2 FIXED Kmin 0.2 FITTED Kmin M = 30

M = 30
0.0 0.0
0 50 100 150 200 250 0 50 100 150 200 250
KJC [MPa √m] KJC [MPa √m]

Figure 22. Failure probability diagram for 12.5 mm thick specimens at T =

-60°C.
22NiMoCr37 T = -600C B = 25 mm
1.4 1.4

1.2 1.2 95 %
FIXED Kmin 95 %
FITTED Kmin
1/ 4

1.0 1.0
1/ 4

5%





f

 ln 1 
 1−P 

0.8 0.8
f
 1− P

5%
 ln 1

0.6 0.6
CLEAVAGE CLEAVAGE


0.4 DUCTILE 0.4 DUCTILE

K0 = 158 MPa√m K0 = 160 MPa√m
0.2 Kmin = 20 MPa√m 0.2 Kmin = 42 MPa√m

0.0 0.0
0 50 100 150 200 250 0 50 100 150 200 250
KJC [MPa √m] KJC [MPa √m]

Figure 23. Failure probability diagram for 25 mm thick specimens at T =

-60°C.
22NiMoCr37 T = -600C B = 50 mm
1.4 1.4
CLEAVAGE
95 %
1.2 DUCTILE 1.2 95 %
K0 = 195 MPa√m FITTED Kmin
 1/ 4

1.0 Kmin = 20 MPa√m 1.0

1/ 4




5%
f

 ln 1 
 1−P 

0.8 0.8
f
 1− P
 ln 1

5%
0.6 0.6
CLEAVAGE


0.4 0.4 DUCTILE

FIXED Kmin K0 = 193 MPa√m
0.2 0.2 Kmin = -5 MPa√m

0.0 0.0
0 50 100 150 200 250 300 0 50 100 150 200 250 300
KJC [MPa √m] KJC [MPa √m]

Figure 24. Failure probability diagram for 50 mm thick specimens at T =

-60°C.

32
Even though the scatter and specimen measuring capacity appear to be
according to the master curve assumptions, the same is not the case for the
expected size effect. There is a decrease in K0 going from 12.5 mm thickness to
25 mm, but the 50 mm thick specimens show a clearly higher toughness than
the smaller specimens. This is not in accordance with weakest link behaviour,
nor is it explainable by constraint speculations. The most logical reason for the
behaviour is a macroscopic material variability. To check this possibility, GKSS
manufactured and tested 30 additional 12.5 mm thick specimens, made from the
broken 50 mm thick specimen halves. These “extra” results are presented in Fig.
25. The K0 for the “extra” results is significantly higher than for the original
25.5 mm thick specimens. It is also higher than the K0 for the 50 mm thick
specimens. Thus the result tends to confirm the existence of a macroscopic
material variability. The measuring capacity of the specimens appears again to
be better than M = 30 indicates.

22NiMoCr37 (2T) (1/2 T specimens made from

T = -600C B = 12.5 mm broken 2 T specimen halves)
1.4 1.4
FIXED Kmin FITTED Kmin
1.2 1.2
95 % 95 %
 1/ 4

1.0 1.0
 1/ 4

5%
5%
f
 ln 1 
 1− P 

f
 ln 1 
 1−P 

0.8 0.8

CLEAVAGE CLEAVAGE
0.6 0.6
DUCTILE DUCTILE


0.4 K0 = 201 MPa√m 0.4 K0 = 202 MPa√m

Kmin = 20 MPa√m Kmin = 2 MPa√m
0.2 0.2
M = 30 M = 30
0.0 0.0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
K JC [MPa √m] KJC [MPa √m]

Figure 25. Failure probability diagram for 12.5 mm thick specimens at T =

-60°C, made from broken 50 mm thick specimen halves.

To further investigate the origin of the inhomogeneity, the data was divided into
two sets, one set containing data from sections SX2 and SX10 and one set
containing data form section SX9. SX10 is the same section as used for the
original 12.5 mm and 25 mm thick specimens. SX2 was used for 50 mm thick
specimens tested at several different temperatures, not showing “anomalous”
behaviour. The results are presented in Fig. 26 for the 50 mm thick specimens
and Fig. 27 for the “extra” 12.5 mm specimens.

33
 22NiMoCr37 T = -600C B = 50 mm
PLATES SX 2 & SX 10
1.4
95 %
1.2
FIXED Kmin PLATE SX 10 USED FOR
 1/ 4

1.0 BASIC 1/2 T AND 1 T





5%
SPECIMENS
f
 1−P

0.8
 ln 1

0.6


CLEAVAGE
0.4 DUCTILE 22NiMoCr37 T = -600C B = 50 mm
K0 = 143 MPa√m PLATE SX 9
0.2
Kmin = 20 MPa√m
1.4
0.0 CLEAVAGE
0 50 100 150 200 250 1.2 DUCTILE 95 %
KJC [MPa √m] K0 = 203 MPa√m
 1/ 4

1.0 Kmin = 20 MPa√m




f
 1−P

0.8
 ln 1

5%
0.6


PLATE SX 9 USED FOR

0.4
MAJORITY OF 2 T
FIXED Kmin
SPECIMENS 0.2

0.0
0 50 100 150 200 250 300
KJC [MPa √m]

Figure 26. Failure probability diagram for 50 mm thick specimens at T =

-60°C, ordered by plate.

22NiMoCr37 (2T) T = -600C B = 12.5 mm

PLATES SX 2 & SX 10
1.4
FIXED Kmin
1.2
95 %
PLATE SX 10 USED FOR
 1/ 4

1.0 BASIC 1/2 T AND 1 T

f
 ln 1 
 1−P 

0.8 SPECIMENS
5% CLEAVAGE
0.6
DUCTILE


0.4 K0 = 194 MPa√m 22NiMoCr37 (2T) T = -600C B = 12.5 mm

Kmin = 20 MPa√m PLATE SX 9
0.2
1.4
M = 30
0.0 FIXED Kmin
0 50 100 150 200 250 300 350 400 1.2
95 %
KJC [MPa √ m]
 1/ 4

1.0
5%
f
 ln 1 
 1−P 

0.8

0.6 CLEAVAGE


DUCTILE
PLATE SX 9 USED FOR 0.4 K0 = 219 MPa√m
MAJORITY OF 2 T Kmin = 20 MPa√m
0.2
SPECIMENS M = 30
0.0
0 50 100 150 200 250 300 350 400
KJC [MPa √m]

Figure 27. Failure probability diagram for 12.5 mm thick specimens at T =

-60°C, made from broken 50 mm thick specimen halves, ordered by plate.

34
The trends are clear. The toughness of section SX9 is approximately 20% higher
than for the neighbouring sections SX2 and SX10. This result indicates that also
the other 50 mm thick specimen data sets may be affected by material
variability. Finding this kind of a moderate material variability in such a big
forging, is by no means unexpected. Actually, the found material variability is
almost surprisingly small. The effect of the material variability on T 0 is further
studied in the synthesis analysis.

4.5 T = -40°C

Three smallest specimen sizes were tested at -40°C and the non-size-adjusted
results are presented in Figs 28–30. Only for the 25 mm thick specimens, there
were 25% specimens showing lack of single initiation sites. Thus the basic
master curve expression should work. All sets follow the predicted distribution
quite well. At this temperature also the 25 mm thick specimens start to reach
fracture toughness values in excess of the size requirement (Eq. 7). A deviation
from the straight line behaviour does occur well beyond the M = 30 line, thus
providing additional validation for the standard size requirement. The data show
also a clear size effect. The size effect appears larger than for the lower
temperatures, but this is at least partly apparent. The size effect should scale
with (K0-Kmin) which indicates larger size effect at higher toughness levels. The
test results seem to verify this trend.

22NiMoCr37 T = -400C B = 12.5 mm

1.4 1.4
95 % 95 %
1.2 1.2
1/ 4

1.0 1.0
1/ 4

FIXED Kmin
f

 ln 1 
 1− P 

FITTED Kmin





0.8 0.8
f

5%
 1− P

5%
 ln 1

0.6 CLEAVAGE 0.6 CLEAVAGE



DUCTILE DUCTILE


0.4 K0 = 289 MPa√m 0.4 K0 = 275 MPa√m

Kmin = 20 MPa√m Kmin = 3 MPa√m
0.2 0.2
M = 30 M = 30
0.0 0.0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
K JC [MPa √m] KJC [MPa √m]

Figure 28. Failure probability diagram for 12.5 mm thick specimens at T =

-40°C.

35
 22NiMoCr37 T = -400C B = 25 mm
1.4 1.4
CLEAVAGE CLEAVAGE
1.2 DUCTILE 95 % 1.2 DUCTILE 95 %
K0 = 231 MPa√m K0 = 231 MPa√m
1 4

1.0 Kmin = 20 MPa√m 1.0

1/ 4
Kmin = 15 MPa√m
/







 ln 1 
 1−P 
0.8 0.8
f

5%
 1−P

5%

f
 ln 1

0.6 0.6



0.4 0.4
FITTED Kmin
FIXED Kmin
0.2 0.2
M = 30 M = 30
0.0 0.0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
KJC [MPa √m] KJC [MPa √m]

Figure 29. Failure probability diagram for 25 mm thick specimens at T =

-40°C.

22NiMoCr37 T = -400C B = 50 mm
1.4 1.4

1.2 1.2 95 %
95 %
 1/ 4

1.0 FIXED Kmin 1.0 FITTED Kmin

1 4
/
f
 ln 1 
 1−P 


 ln 1 
 1−P 

0.8 0.8
5%
f

5%
0.6 CLEAVAGE 0.6 CLEAVAGE
DUCTILE


DUCTILE


0.4 K0 = 181 MPa√m 0.4 K0 = 183 MPa√m

Kmin = 20 MPa√m Kmin = 34 MPa√m
0.2 0.2

0.0 0.0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
KJC [MPa √m] KJC [MPa √m]

Figure 30. Failure probability diagram for 50 mm thick specimens at T =

-40°C.

4.6 T = -20°C

All four specimen sizes were tested at -20°C and the non-size-adjusted results
are presented in Figs 31–34. The basic master curve expression should work for
all specimen sizes. At this temperature, nearly all of the 12.5 mm thick
specimens (Fig. 31) reach fracture toughness values in excess of the size
requirement (all but 2) and also the number of non-failures (ductile end-of-tests)
is considerable. In this case, the fitted Kmin estimate is meaningless, and also the
K0 estimate, even for the fixed K0 case is unreliable. It does, however, appear
that loss of constraint becomes significant only above the M = 30 toughness
criterion. The 25 mm and 50 mm thick specimens (Figs 32 and 33) follow the

36
master curve assumptions nicely and provide also additional verification for the
validity of the standard size requirement.

The 100 mm thick specimens (Fig. 34), indicate a higher Kmin value, but also the
standard value is possible considering the confidence of the Kmin estimate.

22NiMoCr37 T = -200C B = 12.5 mm

1.4 1.4 95 %
FIXED Kmin 95 % FITTED Kmin
1.2 1.2
1/ 4

1.0 1.0
1/ 4










0.8 0.8
f
 1− P

f
 ln 1

 1− P
 ln 1

0.6 0.6


5% CLEAVAGE CLEAVAGE


0.4 0.4 5%
DUCTILE DUCTILE
K0 = 387 MPa√m K0 = 435 MPa√m
0.2 0.2
M = 30 Kmin = 20 MPa√m M = 30 Kmin = -483 MPa√m
0.0 0.0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
K JC [MPa √m] KJC [MPa √m]

Figure 31. Failure probability diagram for 12.5 mm thick specimens at T =

-20°C.

22NiMoCr37 T = -200C B = 25 mm
1.4 1.4
95 %
95 %
1.2 FIXED Kmin 1.2
FITTED Kmin
1/ 4

1.0 1.0
1/ 4

5% 5%
f

 ln 1 
 1− P 






0.8 0.8
f
 1− P
 ln 1

0.6 CLEAVAGE 0.6

CLEAVAGE


DUCTILE


0.4 0.4 DUCTILE

K0 = 342 MPa√m K0 = 330 MPa√m
0.2 Kmin = 20 MPa√m 0.2 Kmin = 39 MPa√m
M = 30 M = 30
0.0 0.0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
K JC [MPa √m] KJC [MPa √m]

Figure 32. Failure probability diagram for 25 mm thick specimens at T =

-20°C.

37
 22NiMoCr37 T = -200C B = 50 mm
1.4 1.4
CLEAVAGE CLEAVAGE
DUCTILE 95 % DUCTILE
1.2 1.2 95 %
K0 = 257 MPa√m K0 = 255 MPa√m
1 4

1.0 Kmin = 20 MPa√m 1.0 Kmin = -6 MPa√m

 1/ 4
/

5% 5%




f
 ln 1 
 1−P 
0.8 0.8
f
 1−P
 ln 1

0.6 0.6



0.4 FIXED Kmin 0.4 FITTED Kmin
0.2 0.2
M = 30 M = 30
0.0 0.0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
KJC [MPa √m] KJC [MPa √m]

Figure 33. Failure probability diagram for 50 mm thick specimens at T =

-20°C.

22NiMoCr37 T = -200C B = 100 mm

1.4 1.4

1.2 95 % 1.2 95 %

FIXED Kmin
1 4

1.0 1.0 FITTED Kmin

1/ 4
/

5%
5%




f

 ln 1 
 1−P 

0.8 0.8
f
 1−P
 ln 1

0.6 CLEAVAGE 0.6 CLEAVAGE



DUCTILE DUCTILE


0.4 0.4 K0 = 205 MPa√m

K0 = 201 MPa√m
0.2 Kmin = 20 MPa√m 0.2 Kmin = 108 MPa√m

0.0 0.0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
K JC [MPa √m] KJC [MPa √m]

Figure 34. Failure probability diagram for 100 mm thick specimens at T =

-20°C.
Additionally 20, side-grooved, specimens were manufactured and tested at this
temperature. The specimens had standard 10% side-grooves on both sides and
the results are presented in Fig. 35. The side grooved specimens provide a 7%
lower value than the non-side-grooved specimens (Fig. 32). The expected
standard deviation (Eq. 6) of the respective K0 estimates are 8.4% (0% SG) and
9.3 (20% SG), which means that the found difference is not statistically
significant. In order to compare the two specimen types in more detail, the 20
non-side-grooved specimens tested outside GKSS were compared directly with
the 20 side-grooved specimens tested by the same laboratory. Both sets of data
were ordered by rank and each of the corresponding data pairs (in principle
corresponding to the same rank probability) were compared. The comparison is
presented in Fig. 36. It should be pointed out that the uncertainty of the

38
comparison is √2-times bigger than for the individual analysis. Thus the
observed difference of 10% (Fig. 36) is not statistically significant. More
revealing is the overall trend in the data. The relation between the two
specimens is linear almost up to ductile maximum load (see Fig. 9) after which
the non-side-grooved values seem to deviate to higher toughness values. I.e. the
both specimens show an identical constraint behaviour up to ductile maximum
load. Finally, also the tearing resistance of the two specimen types were
compared (Fig. 37). Based on a power law fit to the data, also here, the side-
grooved specimens indicate a lower toughness than the non-side-grooved ones.
The difference is clearer for the post maximum load values, but there appears to
be a slight difference also for lower toughness values. Thus, the ductile tearing
results are in agreement with the brittle fracture behaviour. Side-grooves
produce a slightly more conservative result.

22NiMoCr37 T = -200C B = 25 mm 20 % SG
1.4 1.4
95 % 95 %
1.2 FIXED Kmin 1.2
FITTED Kmin
1/ 4

1.0 1.0
1/ 4

5%





5%


 ln 1 
 1−P 

0.8 0.8
f
 1−P

f
 ln 1

0.6 CLEAVAGE 0.6 CLEAVAGE



DUCTILE DUCTILE


0.4 K0 = 318 MPa√m 0.4 K0 = 318 MPa√m

Kmin = 20 MPa√m Kmin = -18 MPa√m
0.2 0.2
M = 30 M = 30
0.0 0.0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
KJC [MPa √m] K JC [MPa √m]

Figure 35. Failure probability diagram for 25 mm thick specimens, with 20%
side-grooving, at T = -20°C.

39
 22NiMoCr37 T = -200C B = 25 mm

500
KJC (20 % SG) = 0.9⋅ KJC (0 % SG)
KJC (20 % SG) [MPa√ m]

400

300

200

100

0
0 100 200 300 400 500 600
K JC [MPa √ m]

Figure 36. Comparison between fracture toughness of side-grooved and non-

side-grooved 25 mm thick specimens at -20°C.

22NiMoCr37 T = -200C B = 25 mm

600
0.37
KJ = 456⋅∆a
500
KJ [MPa√m]

0.32
400 KJ = 421⋅∆a

300
0 % SG
200 20 % SG

100

0
0.0 0.5 1.0 1.5 2.0 2.5
∆a [mm]

Figure 37. Comparison between tearing resistance of side-grooved and non-

side-grooved 25 mm thick specimens at -20°C.

4.7 T = -10°C

At -10°C only five 25 mm thick specimens were tested to check that upper shelf
behaviour would not be encountered earlier than predicted. The results are
presented in Fig. 38. No valid results were obtained, so a master curve
interpretation of the results was not pursued. Only one of the five specimens did
not cleave, so it was decided to continue to increase the testing temperature still.

40
 22NiMoCr37 T = -100C B = 25 mm

1.4
CLEAVAGE
1.2 DUCTILE
K0 = -- MPa√m
1/ 4

1.0
Kmin = -- MPa√m
f

 ln 1 
 1−P 

0.8

0.6


0.4

0.2
M = 30
0.0
0 100 200 300 400 500 600 700 800
KJC [MPa √m]

Figure 38. Failure probability diagram for 25 mm thick specimens at T =

-10°C.

4.8 T = 0°C

All four specimen sizes were tested at 0°C. The non-size-adjusted results are
presented in Figs 39–42. As expected, the 12.5 mm and 25 mm thick specimens
did not provide any results fulfilling the specimen size criterion (Figs 39 and
40). Thus, for these specimens a master curve analysis was not applicable. More
than half of the 50 mm thick specimens (Fig. 41) and all 100 mm specimens
(Fig. 42) fulfilled the size criterion, so for these cases the master curve
expression should work. The 50 mm specimens seem to loose constraint above
M = 30 whereas the 100 mm specimens all follow nicely the predicted
behaviour. A size effect is also clearly present. In order to get an understanding
of the amount of loss of constraint for the smaller specimens, a prediction based
on the 100 mm specimen behaviour was made for these sets (Figs 39 and 40).
Clearly, the small specimens have been affected by a loss of constraint, so data
sets where all specimens violate the size criterion should obviously not be used
for determining the master curve.

41
 22NiMoCr37 T = 00C B = 12.5 mm
1.4
CLEAVAGE
DUCTILE
1.2
*
K0 = 546 MPa√m
 1/ 4

1.0 Kmin = 20 MPa√m

f
 ln 1 
 1−P 

0.8

0.6


M = 30
0.4

0.2 *Prediction from B = 100 mm

0.0
0 100 200 300 400 500 600
KJC [MPa √m]

Figure 39. Failure probability diagram for 12.5 mm thick specimens at T = 0°C.

22NiMoCr37 T = 00C B = 25 mm
1.4
CLEAVAGE
1.2 DUCTILE
*
K0 = 463 MPa√m
1.0
 1/ 4

Kmin = 20 MPa√m




0.8
f
 1−P
 ln 1

0.6


0.4

0.2
M = 30 *Prediction from B = 100 mm
0.0
0 100 200 300 400 500 600 700 800
KJC [MPa √m]

Figure 40. Failure probability diagram for 25 mm thick specimens at T = 0°C.

42
 22NiMoCr37 T = 00C B = 50 mm
1.4 1.4

FIXED Kmin FITTED Kmin

1.2 1.2
95 %
 1/ 4

1.0 95 % 1.0

1/ 4




5%
0.8 0.8 5%

f

 ln 1 
 1−P 
f
 1− P
 ln 1

0.6 CLEAVAGE 0.6 CLEAVAGE

DUCTILE


DUCTILE


0.4 K0 = 407 MPa√m 0.4 K0 = 408 MPa√m
Kmin = 20 MPa√m Kmin = -16 MPa√m
0.2 0.2
M = 30 M = 30
0.0 0.0
0 100 200 300 400 500 600 700 800 900 0 100 200 300 400 500 600 700 800 900
KJC [MPa √m] KJC [MPa √m]

Figure 41. Failure probability diagram for 50 mm thick specimens at T = 0°C.

22NiMoCr37 T = 00C B = 100 mm

1.4 1.4
FITTED Kmin
1.2 FIXED Kmin 1.2
95 % 95 %
1/ 4

1.0 1.0
1/ 4










0.8 CLEAVAGE 0.8

f
 1−P

CLEAVAGE
f
 ln 1

 1− P

5%
 ln 1

5% DUCTILE DUCTILE
0.6 K0 = 333 MPa√m 0.6
K0 = 331 MPa√m


Kmin = 20 MPa√m Kmin = -24 MPa√m

0.4 0.4

0.2 0.2
M = 30 M = 30
0.0 0.0
0 100 200 300 400 500 600 0 100 200 300 400 500 600
K JC [MPa √m] KJC [MPa √m]

Figure 42. Failure probability diagram for 100 mm thick specimens at T = 0°C.

4.9 T = +20°C

The final test temperature was +20°C where the three largest specimen sizes
were tested. The non-size-adjusted results are presented in Figs 43–45. Neither
25 mm nor 50 mm specimens were able to provide any values fulfilling the size
requirement and a master curve analysis was therefore not applicable. The 100
mm thick specimens had all but two values in excess of the size requirement.
Thus, the fitted Kmin estimate is meaningless, and also the K0 estimate, even for
the fixed K0 case is unreliable. The results do, however, imply that loss of
constraint becomes significant only above the M = 30 toughness criterion.

43
The results show also another significant aspect, related to the existence of a
“upper shelf” transition temperature. The results show that cleavage fracture
initiation is possible at very high K Jc-values and at high temperatures. No
absolute “upper shelf” transition temperature was found. Thus, the master curve
assumption that the brittle to ductile transition is nothing else than a
combination of two separate fracture mechanisms is supported by the present
results. The brittle to ductile transition is not a true material property. It is
always related to the structural size. A large structure, allowing for much ductile
crack growth will have a higher transition temperature than a smaller structure
of the same material and this is true even if the constraint of the structures is the
same. Any definition of an upper shelf transition temperature should be based
on a constant specimen geometry and size. And it should be recognised that this
transition temperature will be different for a real structure.

22NiMoCr37 T = +200C B = 25 mm
1.4
CLEAVAGE
1.2
DUCTILE
K0 = -- MPa√m
1/ 4

1.0
Kmin = -- MPa√m
f

 ln 1 
 1−P 

0.8

0.6


0.4

0.2
M = 30
0.0
0 100 200 300 400 500 600 700 800 900
KJC [MPa √m]

Figure 43. Failure probability diagram for 25 mm thick specimens at T =

+20°C.

44
 22NiMoCr37 T = +200C B = 50 mm
1.4

1.2 CLEAVAGE
DUCTILE
1.0 K0 = -- MPa√m
1/ 4

Kmin = -- MPa√m
f

 ln 1 
 1−P 

0.8

0.6


0.4

0.2
M = 30
0.0
0 200 400 600 800 1000 1200
KJC [MPa √m]

Figure 44. Failure probability diagram for 50 mm thick specimens at T =

+20°C.

22NiMoCr37 T = +200C B = 100 mm

1.4 CLEAVAGE 1.4 95 %
95 %
DUCTILE FITTED Kmin
1.2 1.2
K0 = 927 MPa√m
Kmin = 20 MPa√m
1/ 4

1.0 1.0
1/ 4





f

 ln 1 
 1−P 

0.8 0.8
f
 1− P
 ln 1

5%
0.6 5% 0.6
CLEAVAGE


0.4 M = 30 0.4 DUCTILE

FIXED Kmin K0 = 737 MPa√m
0.2 0.2 Kmin = -130 MPa√m
M = 30
0.0 0.0
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
KJC [MPa √m] KJC [MPa √m]

Figure 45. Failure probability diagram for 100 mm thick specimens at T =

+20°C.

45
 5. Synthesis analysis
The synthesis analysis will make an assessment, based on the master curve
predictions, of the lower shelf behaviour of the initiation sites, the size effect of
K0, the temperature dependence of the master curve and the validity of a fixed
Kmin = 20 MPa√m assumption. Finally, an overall multi-temperature master
curve analysis is made and recommendations for the applicability of the master
curve are given.

5.1 Lower shelf behaviour

The lower shelf behaviour is studied by examining the percentage of specimens

not showing single initiation sites on the fracture surface (Fig. 13). Two ways of
presentation was selected: 1)as a function test temperature and 2)as a function K0.
The results are presented in Figs 46 a and b. When plotting the results against
temperature there seems to be a clear size effect on the lower shelf behaviour
(large specimens are more prone to show lower shelf behaviour than small
specimens) and also the scatter is quite large. Plotting the results against K 0
removes some of the size effect and decreases the scatter.

46
 22NiMoCr37
100

B = 12.5 mm
80 B = 25 mm
% Lower Shelf

B = 50 mm
B = 100 mm
60

40

20

0
-180 -160 -140 -120 -100 -80 -60 -40 -20 0
a) 0
T [ C]
22NiMoCr37
100

80 B = 12.5 mm
B = 25 mm
% Lower Shelf

B = 50 mm
60 B = 100 mm

40

20

0
0 50 100 150 200 250 300 350 400
b) K0 [MPa√m]

Figure 46. Percentage of specimens not showing single initiation sites as a

function of a) temperature and b) K0.

From the fracture toughness results it was clear that only the data sets showing
practically 100% lower shelf behaviour on the fracture surface, also showed the
expected scatter and lack of size effect. This is fully in line with the master
curve assumptions. Close to the lower shelf, initiation will for some specimens
be much easier than propagation and these specimens will show lower shelf type
fracture surfaces. For other specimens propagation will be easier than initiation
and these specimens will show single initiation sites on the fracture surface.
Their overall distribution will however follow the standard master curve
distribution, which includes the conditional propagation criterion (leading to
non-zero Kmin). Only when in all cases, the probability of propagation controls
the fracture event, the lower shelf master curve distribution function will come

47
into effect. This means that fracture surface appearance can be used to choose
distribution function. If any specimens at a specific temperature show single
initiation sites, the scatter is likely to follow the standard master curve. Only if
all specimens show lower shelf type fracture surfaces, the lower shelf master
curve may be more appropriate (providing a sufficient number of tests have
been performed N >> 6). If in doubt, the standard master curve distribution
should be assumed.

5.2 Size effect of K0

All the MML (fixed Kmin) K0 estimates were ordered by test temperature and
plotted against specimen thickness and compared with the master curve size
effect predictions (Fig. 47). In all cases the validity of the master curve size
effect (Eq. 8) is confirmed quite clearly. This is true even for the lower shelf
temperature (-154°C). I.e. even the standard master curve predicts negligible
size effects on the lower shelf. Even more important is the finding that there is
no trend for the smaller specimens to show a significantly larger size effect than
predicted by the master curve. Even for the smallest specimens tested at -40°C
and -20°C the predictions are within 10%, i.e. of the same order as the
theoretical accuracy of the MML Ko estimate. Note that the 12.5 mm thick
specimens at -20°C contained only 2 results fulfilling the size requirement. As
many as 28 specimens were censored to correspond to the M = 30 value.

It appears that, at least for this type of materials, a size criterion of M = 30 is

actually sufficient even for single specimen values as defined e.g. in ASTM
E1820 or ESIS P2. Anyway, the results provide a very strong confirmation of
the validity of the standard master curve size requirement of M = 30.

48
 22NiMoCr37
600

MASTER CURVE SIZE EFFECT

500
PREDICTIONS
K0 [MPa√m]

400
0
-154 C
300 -91 C
0

0
-60 C
200 0
-40 C
0
-20 C
100
0
0C
0
0 20 40 60 80 100
B [mm]

Figure 47. Size effect of MML estimated K0 values compared with master curve
predictions of the size effect.

5.3 Temperature dependence of K0

The temperature dependence of the size adjusted (B0 = 25 mm) K0 values was
compared with the standard master curve assumption. Two means of
comparison were used. A direct comparison in terms of K0 which is presented in
Fig. 48 and an indirect comparison in terms of the calculated standard T0 values
which are presented in Figure 49. Into the comparison were included also the
estimates corresponding to the separate plate SX9 and plates SX2 and SX10.
This was made in order to include the effect of the “extreme” variability on the
scatter of T0.

With the exception of plate SX9 (at -60°C), all K0 results show a very consistent
behaviour. No systematic differences between different size specimens are
present. A best fit of the same exponential form as the standard master curve
provided a result very close to the standard master curve (Fig. 48). The main
difference (proportionally) occurs on the lower shelf. For this material the
master curve expression should therefore be slightly modified for the lower
shelf. The standard master curve is, however, not meant to be applied blindly

49
down to lower shelf temperatures. If the lower shelf fracture toughness needs to
be assessed, tests corresponding to lower shelf temperatures should be
performed. The overall standard deviation of the T0 estimates (not including the
lower shelf results, but including plate SX9) is 6.7°C. Since the theoretical
average standard deviation of the T0 estimate is approximately 4°C, it means
that the material related scatter is only 5°C, which, considering the amount of
material sampled, is very little.

22NiMoCr37
500
450 B = 12.5 mm STANDARD
B = 25 mm MASTER CURVE
400
[MPa√m]

B = 50 mm
350
B = 100 mm
300 B0 = 25 mm
BEST FIT
250
25mm

TO DATA
K0

200
150
100
50
0
-160 -140 -120 -100 -80 -60 -40 -20 0 20
0
T [ C]

Figure 48. Temperature dependence of size adjusted K0 values compared with

standard master curve temperature dependence.

50
 22NiMoCr37
-30

-40 Lower Shelf B = 12.5 mm

Not Included B = 25 mm
-50
B = 50 mm
-60 B = 100 mm
T0 [ C]

B0 = 25 mm
0

-70

-80 _
0
T0 = -90 C
-90
σT = 6.7 C
0
0
-100

-110
-160 -140 -120 -100 -80 -60 -40 -20 0 20
0
T [ C]

Figure 49. Comparison of T0 estimates from the individual data sets.

5.4 Validity of fixed Kmin

The validity of the assumption of a fixed value for the limiting fracture
toughness (Kmin = 20 MPa√m) was investigated by plotting the individual MML
Kmin estimates as a function of temperature (Fig. 50). The scatter in Kmin
increases with temperature (K0), but this is as expected based on Fig. 4. The
results were compared with the theoretical 5% and 95% confidence bounds for
the MML estimate (Fig. 4) assuming a true value of 20 MPa√m. No trends with
respect to specimen size nor temperature are visible. Thus, at least for this
material, the assumption of a constant temperature independent Kmin value close
to 20 MPa√m is verified.

51
 22NiMoCr37
120

100 B = 12.5 mm
B = 25 mm
80
Kmin [MPa√m]

B = 50 mm
60 B = 100 mm

40

20

0
THEORETICAL
-20 5 % & 95 %
ESTIMATES
-40
-160 -140 -120 -100 -80 -60 -40 -20 0 20
0
T [ C]

Figure 50. Temperature dependence and scatter of MML estimate of Kmin

compared to 5% and 95% confidence bounds for fixed Kmin of 20 MPa√m

5.5 Multi-temperature master curve analysis

As a final check of the applicability of the master curve method, the different
size specimen data were analysed by the multi-temperature T0 MML algorithm
(Eq. 10). The analysis results are presented in Figs 51–54a and b. Two different
fits were made one applying all data (Figs 51–54 a) and one applying only data
in the range -50°C ≤ T - T0 ≤ +100°C (Figs 51–54b). The T0 estimates applying
all data vary from -85°C down to -97°C. Applying only the central part of the
data decreases the scatter in T0 a little (-87°C…-97°C). The standard deviation
in this case is 4°C, which is well in line with the expected material variability.
There is an indication that the highest test temperature has a higher fracture
toughness than predicted by the master curve. Thus, too low and too high test
temperatures should be avoided for the estimation of T0.

Based on the results, and previous theoretical estimates (Fig. 3), it seems
advisable to limit the master curve T 0 estimation to testing temperatures in the
range -50°C ≤ T - T0 ≤ +50°C. Testing should include several temperatures

52
(preferably more than three), in order to minimise any effects from a possible
deviation from the assumed temperature dependence.

22NiMoCr37 σY = 467 MPa B = 100 mm N = 55 22NiMoCr37 σY = 467 MPa B = 100 mm N = 40

700
1800
CLEAVAGE 600 CLEAVAGE
1600
DUCTILE DUCTILE
1400 0 0

KJC [MPa√m]
KJC [MPa√m]

T0 = -97 C 500 T0 = -90 C

B0 = 25 mm B0 = 25 mm
1200
400
1000
M = 30 95 %
800 300

600 95 % 200
400 5%
100
200 5%

0 0
-25 0 25 50 75 100 125 -50 -25 0 25 50 75 100
a) T - T 0 [ C]
0 b) 0
T - T 0 [ C]

Figure 51. Multi-temperature MML T0 analysis of 100 mm thick specimen data.

a) using all data, b) using data in the range -50°C ≤ T0 ≤ +100°C.

22NiMoCr37 σY = 467 MPa B = 50 mm N =210 22NiMoCr37 σY = 467 MPa B = 50 mm N =150

1400 1000

1200 CLEAVAGE CLEAVAGE

DUCTILE 800 DUCTILE
KJC [MPa√m]

0
KJC [MPa√m]

0
1000 T0 = -96 C T0 = -97 C
B0 = 25 mm B0 = 25 mm
800 600
M = 30

600 M = 30 95 %
400 95 %

400

5% 200
200
5%

0 0
-75 -50 -25 0 25 50 75 100 125
a) 0 b)
-50 -25 0 25
0
50 75 100
T - T 0 [ C] T - T 0 [ C]

Figure 52. Multi-temperature MML T0 analysis of 50 mm thick specimen data.

a) using all data, b) using data in the range -50°C ≤ T0 ≤ +100°C.

53
 22NiMoCr37 σY = 467 MPa B = 25 mm N = 224 22NiMoCr37 σY = 467 MPa B = 25 mm N = 180

800 800
CLEAVAGE CLEAVAGE
700 DUCTILE 700 DUCTILE
KJC [MPa√m]

KJC [MPa√m]
0
600 T0 = -89 C T0 = -91 C
600
B0 = 25 mm B0 = 25 mm
500 500
400 M = 30 95 % 400 95 %
M = 30
300 300
200 5% 200
5%
100 100
0 0
-100 -50 0 50 100 -50 -25 0 25 50 75 100
a) 0
T - T 0 [ C] b) 0
T - T 0 [ C]

Figure 53. Multi-temperature MML T0 analysis of 25 mm thick specimen data.

a) using all data, b) using data in the range -50°C ≤ T0 ≤ +100°C.

22NiMoCr37 σY = 467 MPa B = 12.5 mm N = 268 22NiMoCr37 σY = 467 MPa B = 12.5 mm N = 238

500 CLEAVAGE 500 CLEAVAGE

DUCTILE DUCTILE
KJC [MPa√m]

0
KJC [MPa√m]

0
T0 = -85 C T0 = -87 C
400 400
B0 = 25 mm B0 = 25 mm 95 %

300 300
M = 30
M = 30
200 200
5% 5%
95 %
100 100

0 0
a) -100 -50 0
0
50 100
b)
-50 -25 0 25
0
50 75 100
T - T 0 [ C] T - T 0 [ C]

Figure 54. Multi-temperature MML T0 analysis of 12.5 mm thick specimen data.

a) using all data, b) using data in the range -50°C ≤ T0 ≤ +100°C.

54
 6. Summary and conclusions

A large nuclear grade pressure vessel forging 22NiMoCr37 (A508 Cl.3) has
undergone extensive fracture toughness testing. The tests were performed on
standard geometry CT-specimens having thickness 12.5 mm, 25 mm, 50 mm and
100 mm. The a/W-ratio was close to 0.6 for all specimens. One set of specimens
was 20% side-grooved. A total of 757 results fulfilling the ESIS-P2 test method
validity requirements with respect to pre-fatigue crack shape and the ASTM E-
1921 pre-fatigue load, were obtained. The master curve statistical analysis
method was applied extensively on the data, in order to verify the validity of the
method. Based on the analysis the following can be concluded regarding the
validity of the master curve method for this material:

1. The master curve scatter assumption is valid.

2. The master curve size effect assumption is valid.

3. The master curve temperature dependence assumption is valid.

4. The master curve minimum fracture toughness assumption is valid.

5. The master curve (E1921-98) specimen size requirement is valid.

6. The master curve lower shelf behaviour assumption is valid.

7. Lower shelf behaviour can be suspected if no specimens show single

initiation sites on the fracture surface.

8. Determination of T0 should be based on test results in the temperature

range -50°C ≤ T - T0 ≤ +50°C.

9. Testing should include several test temperatures, in order to minimise any

effects from a possible small deviation from the master curve temperature
dependence.

10. If only approximate (lower bound type) information regarding the fracture
toughness is required, the master curve can well be extrapolated outside the
range -50°C ≤ T - T0 ≤ +50°C.

55
11. If an accurate description of the fracture toughness outside this temperature
range is required, tests should preferably be performed at the specific
temperature of interest. The master curve analysis method (excluding the
temperature extrapolation) can be used also in this case for the description
of scatter and size effects.

12. For side-grooved specimens, the proper thickness to be used in connection

with the master curve size adjustment is the specimen nominal thickness
not the net thickness.

13. Single temperature and multi-temperature MML T0 estimation algorithms

yield equivalent estimates.

14. The use of random censoring is validated.

56
 References
1. Wallin K. The scatter in K-results. Engineering Fracture Mechanics,
1984, Vol. 19, No. 6, pp. 1085−1093.

2. Wallin, K. Optimized estimation of the Weibull distribution parameters.

Espoo: Technical Research Centre of Finland, 1989. 17 p. + app. 1 p.
(Research Reports 604).

3. Wallin, K. Recommendations for the application of fracture toughness

data for structural integrity assessments. In: Pugh, C.E., Bass, B. R. &
Keeney, J. A., Proc. of the Joint IAEA/NEA International Specialists’
Meeting on Fracture Mechanics Verification by Large Scale Testing. Oak
Ridge. Tennessee, 26−29 Oct. 1993. Oak Ridge: OECD, 1993.
Pp. 465−494. (NUREG/CP-0131, ORNL/TM-12413 RF).

4. ASTM E1921-97. Standard test method for determination of reference

temperature, T0, for ferritic steels in the transition range. American
Society for Testing and Materials, 1998, Vol. 03.01, pp. 1068–1084.

5. Wallin, K. The size effect in Kic results. Engineering Fracture Mechanics,

1985, Vol. 22, No. 1, pp. 149−163.

6. Wallin, K. Fracture toughness transition curve shape for ferritic structural

steels. In: Teoh, S. H. & Lee, K. H. (Eds.) Proceedings of the Joint
FEFG/ICF International Conference on Fracture Engineering Materials &
Structures. London: Elsevier Applied Science, 1991. Pp. 83−88.

7. Wallin, K. Irradiation damage effects on the fracture toughness transition

curve shape for reactor pressure vessel steels. The International Journal of
Pressure Vessels and Piping, 1993, Vol. 55, No. 1, pp. 61−79.

8. Wallin, K. Validity of small specimen fracture toughness estimates

neglecting constraint corrections. In: Kirk, M. & Bakker, A. (Eds.).
Constraint effects in fracture: Theory and applications. Philadelphia:
American Society for Testing and Materials, 1995. 19 p. (ASTM STP
1244).

57
9. Wallin, K. Statistical aspects of constraint with emphasis on testing and
analysis of laboratory specimens in the transition region. Constraint
effects in fracture. In: Hacket, E. M., Schwalbe, K.-H. & Dodds, R. H. Jr.
(Eds.). Constraint Effects in Fracture. Indianapolis, USA, 8−9 May 1991.
Philadelphia: American Society for Testing and Materials, 1993. Pp.
264−288. (ASTM STP 1171).

10. Wallin, K. Macroscopic nature of brittle fracture. Journal de Physique,

1993, Vol. 3, No. IV, pp. 575−584.

11. Heerens, J. SM&T Project: Fracture toughness of steel in the ductile to

brittle transition regime. Final report, Part I. GKSS Forschungszentrum,
1998. To be published.

58